Research map / Large language models / Agents, code and math
Retro-Expert: Collaborative Reasoning for Interpretable Retrosynthesis
Xinyi Li, Sai Wang, Yutian Lin, Yu Wu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| RL with KGPO instead of SFT with CoT (ablation) | Qwen2.5-7B-Instruct with decision space · USPTO-50K | Top-1 accuracy (%): 48.6 -> 75.7 | Improved |
| replace standard GRPO with KGPO advantage rectification (ablation) | Retro-Expert RL · USPTO-50K | Top-1 accuracy (%): 66.8 -> 75.7 | Improved |
| add reaction-related reward rreact to outcome reward (ablation) | Chem-RM · USPTO-50K | Top-1 accuracy (%): 66.8 -> 72.3 | Improved |
| add structure-related reward rstruct (ablation) | Chem-RM · USPTO-50K | Top-1 accuracy (%): 72.3 -> 75.7 | Improved |
| add decision space from specialized models to general LLMs (ablation) | GPT-4o · USPTO-50K | Top-1 accuracy (%): 0.30 -> 32.85 | Improved |
| add decision space to specialized chemical LLMs (ablation) | T5Chem and nach0 · USPTO-50K | Top-1 accuracy (%): 43.64 -> 27.88 (T5Chem); 30.52 -> 14.60 (nach0) | Hurt |
| add decision space to SFT-trained Qwen2.5-7B-Instruct (ablation) | Qwen2.5-7B-Instruct (SFT) · USPTO-50K | Top-1 accuracy (%): 43.62 -> 58.92 | Improved |
| use Retro-Expert on top of specialized retrosynthesis models | Graph2Edits, UAlign, LocalRetro, GraphRetro, GLN, Retroformer, MHNReact · USPTO-50K | Top-1 accuracy (%): Graph2Edits 67.2 -> 75.7; MHNReact 51.8 -> 60.9 | Mixed |
| use Retro-Expert instead of the best baseline | Qwen2.5-7B-Instruct with KGPO · USPTO-FULL | Top-1 accuracy (%): 49.7 -> 58.4 | Improved |
| use Retro-Expert instead of DeepSeek-R1 and base Qwen2.5-7B | out-of-distribution reactions · ChemBench | Top-1 accuracy (%): 29.33 -> 65.67 | Improved |
| replace GRPO with KGPO (ablation) | candidate rejection behavior · USPTO-50K test | rejection rate of wrong candidates (%): 0.22 -> 39.12 | Improved |
| use KGPO instead of GRPO (ablation) | reasoning quality · USPTO-50K | GPT-4o Mechanism Accuracy (1-5): 3.37 -> 4.41 | Improved |
Closest papers by meaning
- LLM-Augmented Chemical Synthesis and Design Decision Programs (ICML 2025 · similarity 0.91)
- Beyond Chemical QA: Evaluating LLM's Chemical Reasoning with Modular Chemical Operations (NeurIPS 2025 · similarity 0.89)
- Retrieval-Retro: Retrieval-based Inorganic Retrosynthesis with Expert Knowledge (NeurIPS 2024 · similarity 0.89)
- Training a Scientific Reasoning Model for Chemistry (NeurIPS 2025 · similarity 0.88)
- From What to Why: A Multi-Agent System for Evidence-based Chemical Reaction Condition Reasoning (ICLR 2026 · similarity 0.87)
- Advancing Retrosynthesis with Retrieval-Augmented Graph Generation (AAAI 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
